Surfaces influence comfort as well as energy consumption or necessary strength during wheelchair driving. However, information about surfaces is only scarcely available in OpenStreetMap. To increase the density of this information, smartphone sensors combined with crowdsourcing is used. In this work, a concept is introduced, which analyses the surface using vertical acceleration. The focus is placed on an automatically calibrating detection system. With this, a user has no effort calibrating. The calibration is accomplished with a learning algorithm. This algorithm uses areas, from which the surface type is known, to learn. Lastly, the allocation rate of the surface type detector after calibration is evaluated, resulting in a rate between ca. 65% to 97%.